I've done an internet search trying to find the theoretical maximum number of input datasets that the Merge tool will accept. Nothing has turned up. I'm currently using 10.5.1
Unless the Merge documentation specifies a limit, which it does not, I think you should assume that for practical purposes it is unlimited.
My recollection is that the MERGE command of ARC/INFO, from which this tool derives had a limit of 500 coverages, and so I would expect that you would be able to use at least that many feature classes.
If you encounter a limit when using this tool then I think you should revise your question with your evidence for that.